Manifold Filtering Problem Lockheed Martin Jarett Hailes Jonathan Wiersma Richard VanWeelden July 21, 2003
Outline Pod problem description Signal / Observation Model SERP Simulation Parameter Details Results IDEX Implementation Sketch of progress towards proof
Problem Description Pod Sensor Second Sensor
Signal Description : : x h Process WienerD2: t W Stratonovich SDE 3 Dimensional State
Signal Constraints
Signal Implementation Stratonovich SDEIto SDE
Pod Observation Model : basis of plane normal to r r : resting position of pod sensor
) Resample Particles 2) Evolve Particles 3) Update Weights given Y k If W(ξ t i ) < ρ W(ξ t j ): )()( )( j t i t j t WW W Prob: 22 )()( )( j t i t i t WW W )( j t W )( i t W 2 )()( i t j t WW Filtering Using SERP
Simulation Parameters Error Function: r = 1 e Signal Estimate
Simulation
Future Directions - Workable explicit solution -Eliminate approximation errors in SERP particle evolve - Use IDEX as filter - More realistic manifolds - Cantilever equations - Enhance signal motion - Damped Harmonic Motion
θ φ Other Observation Model
Filtering with IDEX Goal: prove explicit solutions exist IDEX provides: Faster computation No inherent approximation error
Background From Kouritzin and Remillard (2000): are 1-step nilpotent, h constraint holds
Problem Description Two dimensional manifold in three space
Conditions
Equivalency By Ito’s formula and martingale theory: (1)
Results Conjecture: Φ exists iff (2) where (2)
Sketch of Proof Assuming that Φ exists, then (2) is equivalent to two-step nilpotentcy Apply chain rule Simplify equation
Sketch of Proof Cont. If (2) holds, and all σ m are two step nilpotent, then Φ exists Idea: find satisfying (3) (4) (5) (6) (7)
Sketch of Proof Cont. Let Φ be such that Get dΦ is exact by (3-7) Future: construct (4-6), show they converge